Hispanic Heritage Month is a time to celebrate the culture, history, traditions, and contributions of Hispanic communities. It’s a celebration of where we come from, the stories that shaped us, and the impact we continue to make. It is also a reminder that our heritage is not just history; it’s alive, growing forward by the next generation.
I once had the opportunity to explore a visualization map created by the New York Times. The project was called ‘An American Mosaic’; a visual representation of ethnicities represented county wide throughout the United States. Its great! You can check it out here.
This got me thinking about population changes at the county-level for Hispanic Americans. The United States Census Bureau have specific population surveys that identify citizen ethnicities or specific to Hispanic populations at the county level. One of those surveys that appealed to me was ACS 5-Year Estimate survey B03001; (Hispanic or Latino Origin by Specific Origin). The dataset includes a county-level analysis that shares populations by ethnicities (Hispanic, Non-Hispanic, Mexican, Puerto Rican, and Cuban). The important limitation identified was that it was not a complete breakdown of every Hispanic ethnicity/origin. So with that identified shortfall; I decided to only visualize Hispanic population datapoints and compare those volumes against their respective county totals. Since I was able to identify five historical ready datasets dating back to 2009. I decided that the best thing to visualize would be Hispanic populations over periods of years (across many 5-year survey datasets) for their respective counties. This would give me the evidence needed to answer the following questions:
Where is the Hispanic population changing, and what does that change look like geographically?
Which counties experienced the largest swings of Hispanic population changes?
Do any population cohorts show growth where its total and/or non-Hispanic totals showed declines?
I think its good to note here that the data defined the story I wanted to tell. I originally wanted to call out every identified Hispanic origin; yet through deeper investigation I was not able to identify specific datasets with that level of datapoints. So this is a good lesson where the data being used did not support my initial idea but letting the data drive the story got me a much better project to work on.
With the B03001 dataset I am now able to identify where Hispanic populations are concentrated and where populations are changing. I wanted to uncover insights from different perspectives. I wanted to look at population deltas and percentage changes; this distinction allows me to view use cases where a county can have a large population change without having the largest percentage change, and vice versa.
Designing the Dashboard Around Exploration (using Maps)
I wanted to build a dashboard where I can leverage Tableau’s native maps functionality, and this dataset was the perfect use case to work on that dashboard skill.
By plotting population percentage changes across counties countywide, we can easily identify those counties that are experiencing high magnitudes of growth and/or declines. Along with an interactive map, I wanted to supplement that view with callouts of top county gainers/decliners and parameters to isolate the From and To years of the 5-year survey data. Using the parameters; you can isolate with two survey time periods can result in the population changes I can visualize in the map. I also wanted to add interactivity to the dashboard where I can click county specific marks to expand county profile boxes to view its respective KPIs based off the From/To Years. This action right here tells me automatically that I can accomplish this with Dynamic Zone Visibility.
The below is my underlying design philosophy:
Start broad >> Find something interesting >> Select It >> Investigate Deeper
In summary, the following functionalities were leveraged in the designed dashboard:
- Dynamic Zone Visibility
- KPI boxes
- Maps
- Parameters
- LOD calculations
- Native charts like scatter plots, bar charts
- Formatting design choices like backgrounds and curved corners
Tableau Problem I Had to Solve
When I was building the county profile box (eventually to appear when clicking on a county mark in the map); I wanted to specifically rank the selected county based off population % changes (parameter year driven for variance % change). The below is ultimately what the county profile looks like:
When I click on Dallas County on the map, the profile box above appears using Dynamic Zone Visibility (DZV). This allows the dashboard to surface specific KPIs for the selected county, including Hispanic population totals, year-over-year growth or decline, population share across each 5-year survey, and the county’s ranking relative to all counties in the dataset. This creates an isolated view of the selected county and provides additional context around how it compares nationally.
I then ran into a challenge: accurately displaying the selected county’s rank against more than 3,000 counties, while allowing that rank to dynamically update whenever I selected a different county or changed the From/To Years being compared.
I didn’t want to display all 3,000+ counties just to calculate the ranking, so I needed a way to evaluate every county behind the scenes while only displaying the selected county’s result.
My initial approach was to use Level of Detail (LOD) calculations to handle the ranking. However, that approach broke down once I removed the county dimension and filtered the view down to only the selected county. I then shifted my approach by identifying the selected county and using window calculations to rank it against all other counties.
Finally, I used an LOD calculation to repeat the selected county’s rank across the underlying county records. Because the rank was now consistent across those records, I could remove the county dimension and aggregate the result down to a single value. This allowed me to display the accurate rank of the selected county without having to show all 3,000+ counties in the visualization. This method allowed me to only show the correct ranking once and confirmed that it would dynamically change across various selected counties.
Design Choice: Less is More
Limiting visual noise beyond the headliner (the county map); I can leverage ‘white space’ to elegantly display my insights/visuals cleanly. While leveraging user interaction to reveal more information rather than adding to other static containers (using DZV); I am able to design a less intimidating visualization for users to interact with. The one major lesson I learned that I can carry with me is: Just because the dataset exists doesn’t mean it belongs on the dashboard. This was a pitfall I used to find myself in all the time; not every data point needs a place on the dashboard.
Build with intent and stick to the story you want to tell.
Here are some snapshots of the dashboard I was able to build:
What I Learned From the Build
- Find inspiration and direction in the questions you want to answer. Never build blindly; have a use case and/or question and build towards that answer.
- Its okay for the dataset to change your original idea; the underlying dataset is not perfect; Limitations exist and remember to stick to the story that your dataset can explain.
- User interaction can reduce clutter, making white space an important part of your visualization build.
- A dashboard can be an exploration tool, not just a reporting tool.
Hispanic Heritage Month gives us an opportunity to celebrate culture, history, and community. For me, this project was a way to explore another dimension of that story: where Hispanic communities are growing, changing, and evolving across the United States.
Check out the dashboard here.